agentleFS
Sign inStart free
luciferprosunluciferprosun/AIOA-spArkHAT/skills/evidence-audit/SKILL.md

evidence-audit

Skill2 starsChanged 6 days agoMITView on GitHub

Use when reviewing AIOA spArkHAT NVIDIA replay safety, authority boundaries, or deterministic evidence claims.

6 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

8 rules. 0 are common, 7 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 5 edited versions we read keep these.

  • 7

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use it for questions about the competition trajectory, replay safety, execution authority, fixture/live boundaries, or submission-facing …

    line 25
  2. Distinctive:

    No API key is required.

    line 32
  3. Distinctive:

    Run the bundled read-only evidence-audit wrapper.

    line 39
  4. Distinctive:

    If status is not PASS, preserve the failure and list the failed gates.

    line 42
  5. Distinctive:

    Do not repair, deploy, enable LIVE providers, or change authority boundaries unless the user separately requests a different workflow.

    line 43
  6. Very specific:

    If the agent host exposes a run_script helper, use:

    line 46
  7. Distinctive:

    all LIVE claims emitted by this run are false.

    line 89
  8. Distinctive:

    Do not expose secrets, hidden chain-of-thought, credential paths, or unrelated repository content.

    line 142

“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

AIOA Evidence Audit

Purpose

Perform a deterministic, read-only NVIDIA reviewer audit of an AIOA spArkHAT source checkout. The skill wraps the repository's existing reviewer preflight and emits a compact summary without granting new authority or changing runtime state.

Use it for questions about the competition trajectory, replay safety, execution authority, fixture/live boundaries, or submission-facing evidence claims.

Prerequisites

  • Python 3.12 or newer.
  • An AIOA spArkHAT source checkout.
  • The checkout must contain scripts/nvidia_reviewer_preflight.py.
  • No API key is required.
  • Hosted-provider credentials are intentionally removed from the wrapper's subprocess environment.

Instructions

  1. Confirm the target is an AIOA spArkHAT source checkout.
  2. Run the bundled read-only evidence-audit wrapper.
  3. Parse the JSON result from stdout.
  4. Report only claims supported by that result.
  5. If status is not PASS, preserve the failure and list the failed gates.
  6. Do not repair, deploy, enable LIVE providers, or change authority boundaries unless the user separately requests a different workflow.

If the agent host exposes a run_script helper, use:

run_script("scripts/audit_aioa.py", args=["--repo", "."])

Otherwise, from the repository root run:

python3 skills/evidence-audit/scripts/audit_aioa.py --repo .

Available Scripts

ScriptPurposeArguments
scripts/audit_aioa.pyRuns the deterministic evidence preflight with provider credentials removed and returns a claim-safe JSON summary.--repo PATH, optional --timeout-s SECONDS

Safety contract

This skill is read-only with respect to the repository and competition target.

It must never:

  • enable or call a hosted model provider;
  • load OpenRouter, NVIDIA, AWS, Gemini, xAI, or DeepSeek credentials;
  • perform a production Service Guard effect;
  • rewrite historical FAIL or UNKNOWN evidence;
  • claim 24H_PASS_CLOSED from this deterministic preflight;
  • present TEST_FIXTURE evidence as LIVE;
  • promote DVM or pheromone mechanisms beyond SHADOW;
  • grant execution authority to provider output, critics, memory, or the skill.

The existing deterministic preflight may exercise only disposable local TEST_FIXTURE effects inside an isolated temporary directory. It must not perform an external or production effect, and its temporary artifacts are deleted when the wrapper exits.

Required PASS evidence

A PASS requires all of the following:

  • preflight scope is DETERMINISTIC_TEST_FIXTURE_ONLY;
  • competition evaluation status is PASS;
  • trajectory stage count is 14;
  • duplicate effects equals 0;
  • restart replay dispatches equals 0;
  • competition effect executor is ServiceGuard;
  • provider mode is TEST_FIXTURE;
  • memory mode is TEST_FIXTURE;
  • DVM / pheromone mode is SHADOW;
  • the read-only trajectory sidecar is emitted as ATIF-v1.7;
  • all LIVE claims emitted by this run are false.

For exact claim wording, read references/CLAIM_BOUNDARIES.md.

What a PASS does not prove

A PASS from this skill does not prove LIVE NVIDIA/Nemotron inference, LIVE OpenRouter CPL execution, LIVE CockroachDB, LIVE AWS Non-Zero, a contiguous 24-hour certification, production DVM/pheromone authority, or arbitrary production deployment.

Examples

Review the default checkout:

python3 skills/evidence-audit/scripts/audit_aioa.py --repo .

Review another checkout with a bounded timeout:

python3 skills/evidence-audit/scripts/audit_aioa.py --repo /path/to/aioa --timeout-s 120

Expected successful output includes "status": "PASS", stage count 14, zero duplicate effects, zero restart redispatches, ServiceGuard as executor, and explicit TEST_FIXTURE labels.

Limitations

  • The skill validates only the deterministic reviewer path.
  • It does not merge separate LIVE evidence into its result.
  • It does not certify 24H_PASS_CLOSED.
  • It does not inspect arbitrary production infrastructure.
  • It does not repair failed gates automatically.
  • It does not prove that an external provider was reachable at review time.

Troubleshooting

ErrorLikely causeAction
TARGET_NOT_AIOA_CHECKOUTThe target does not contain the reviewer preflight.Point --repo at the AIOA spArkHAT checkout.
PREFLIGHT_FAILEDA deterministic reviewer gate failed.Preserve the failure; inspect the repository preflight output separately.
PREFLIGHT_OUTPUT_INVALIDThe preflight did not return valid JSON.Treat the review as failed; do not infer PASS.
PREFLIGHT_TIMEOUTThe local deterministic preflight exceeded the allowed runtime.Investigate local test/runtime health; do not switch to a LIVE provider.

Output expectations

Return status, scope, stage count, duplicate effects, restart redispatches, effect executor, provider mode, memory mode, DVM/pheromone mode, ATIF trajectory schema/digest, whether this run validates any LIVE provider, and a concise list of claims still unproven.

Do not expose secrets, hidden chain-of-thought, credential paths, or unrelated repository content.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free